A Theory of Attention. Variations in The Associability

Psychological Review
1975, Vol. 82, No. 4, 276-298
A Theory of Attention: Variations in the Associability

of Stimuli with Reinforcement
N. J. Mackintosh
University of Sussex, Brighton, England
According to theories of selective attention, learning about a stimulus de-
pends on attending to that stimulus; this is represented in two-stage models
by saying that subjects switch in analyzers as well as learning stimulus-
response associations. This assumption, however, is equally well represented
in a formal model by the incorporation of a stimulus-specific learning-rate
parameter, a, into the equations describing changes in the associative
strength of stimuli. Theories of selective attention have also assumed
(a) that subjects learn to attend to and ignore relevant and irrelevant
stimuli (i.e., that a may increase or decrease depending on the correlation
of a stimulus with reinforcement) and (b) that there is an inverse relation-
ship between the probabilities of attending to different stimuli (i.e., that
an increase in a to one stimulus is accompanied by a decrease in a to others).
The first assumption is used to explain the phenomena of acquired dis-
tinctiveness and dimensional transfer, the second those of overshadowing
and blocking. Although the first assumption is justified by the data, the
second is not: Overshadowing and blocking are better explained by the
choice of an appropriate rule for changing a, such that a decreases to stimuli
that signal no change from the probability of reinforcement predicted by
other stimuli.
In experiments on classical conditioning, response. Whatever differences there may

the arrangement of a contingency between be between these two procedures, it is not
a stimulus and a reinforcer typically es- unreasonable to expect some similarity in
tablishes that stimulus as an effective the processes whereby the CS or discrimi-
conditioned stimulus (CS), which reliably native stimulus are established as effective
elicits a conditioned response (CR). In signals. The two stimuli may signal dif-
experiments on instrumental learning, the ferent events—the simple delivery of a
reinforcement of a particular response only reinforcer in one case, and the delivery
in the presence of a specific stimulus typi- of a reinforcer contingent on the occur-
cally establishes that stimulus as an ef- rence of a response in the other—but one
fective discriminative stimulus, which re- may hope that similar laws govern the
liably controls the occurrence of that acquisition of these signaling properties.
My concern in this article, at any rate,
is with the general question of how stimuli
The research reported in this article was sup-
ported by grants from the National Research are established as signals or acquire asso-
Council of Canada and by the United Kingdom ciative strength, and no distinction will be
Science Research Council. I am indebted to made between classical and instrumental
numerous people who have read earlier versions experiments.
of the paper or patiently listened to me expound
some of the ideas it contains. Although I have Virtually all theories of associative learn-
no doubt assimilated ideas from many of them, ing have taken as their starting point the
I am particularly conscious of having incorporated assumption that increments in the asso-
points suggested by A. Baker, A. Dickinson, ciative strength of a stimulus depend upon
G. Hall, E. Macphail, D. Medin, R. A. Rescorla,
and N. S. Sutherland. I am very grateful to some degree of temporal contiguity be-
them all. tween occurrences of that stimulus and of
Inquiries concerning this article should be sent the event it signals. By those who wish
to N. J. Mackintosh, Laboratory of Experimental
Psychology, University of Sussex, Brighton BN1 to express these matters more precisely,
9QG, England. it may be assumed that the magnitude
276
A THEORY OF ATTENTION 277
of increments to the associative strength tioning proceeds faster with stimulus A

of a particular stimulus, A, is some function than with stimulus B, or if subjects, given
of the current value of A's associative conditioning trials with a compound stim-
strength, the simplest version of this as- ulus consisting of A + B, respond strongly
sumption being that such increments are to A but weakly to B, these facts may be
a linearly decreasing function of that value. represented by assigning a high value to
This assumption is expressed formally by «A, and a low value to «B.
the application of linear operators to To this point, the elementary ideas being
changes in associative strength: advanced do not represent any significant
departure from the central assumptions
AF A = 0(\ - FA), (1) of traditional theories of learning, at least
where FA = current associative strength of theories such as those advanced by Hull
of A; 6 is a learning-rate parameter, the (1943), Spence (1936,1956), or Estes (1950,
value of which is determined by the rein- 1959). At least two recently proposed
forcer used within the limits 0 < 6 < 1; models in this tradition, for example, have
and X = limit of F, also dependent on the included a stimulus-specific learning-rate
reinforcer used. parameter such as a (Kendler, 1971;
This may be a starting point, but it is Rescorla & Wagner, 1972).
clearly no more than that. For example,
it does not provide any account of one of THEORIES OF SELECTIVE ATTENTION
the most obvious features of both classical The framework expressed by Equa-
and instrumental conditioning, namely, tion 2 is, in fact, sufficiently general to
the observation that increments in asso- include another group of theories, super-
ciative strength depend on the nature of ficially quite distinct. Theories of selec-
the stimulus. Pavlov (1927) noted that tive attention, of which several versions
a more intense stimulus was a more ef- have been advanced in recent years (for
fective CS than a less intense one, and example, Lovejoy, 1968; Sutherland, 1964;
that in the case of salivary conditioning Sutherland & Mackintosh, 1971; Trabasso
with dogs, stimuli from some modalities, & Bower, 1968; Zeaman & House, 1963),
such as tactile and auditory stimuli, were appear to incorporate some quite radically
more effective CSs than stimuli from other new assumptions about the processes in-
modalities, such as visual or thermal volved in even simple cases of instru-
stimuli. Similarly, the wavelength of light mental learning. Nevertheless, I shall
to which pigeons are exposed is more likely argue that their two central assumptions
to gain control over their food-reinforced can easily be represented as variants on
instrumental responses than is the fre- Equation 2.
quency of a tone sounded from a speaker Theories of selective attention have all
(Foree & LoLordo, 1973). The simplest been formalized as "two-stage" or "chain-
way of representing these and similar ob- ing" models, where the subject is envisaged
servations is to assume that the rate of as first observing or attending to a set of
conditioning to a particular stimulus is stimuli on each trial, and then making an
not just determined by some general overt response determined only by those
learning-rate parameter, such as 6, but is stimuli attended to on that trial. In this
also affected by a stimulus-specific param- form, they seem to differ sharply from the
eter, which may be referred to as a. Thus, "single-stage" type of model characterized
by Equation 2. Yet although the func-
AFA = - FA), (2)
tion of the observing or attentional stage
where 0 < a < 1. It is now possible to may be described informally as that of
represent differences in the rate of condi- initially exposing the subject to a parti-
tioning to stimuli differing in intensity or cular set of discriminative stimuli so that
salience as differences in the value of a those stimuli and no others are available
associated with that stimulus. If condi- for association with the outcome of the
278 N. J. MACKINTOSH
trial, formally it is only this second point-— but in the more interesting case, it is a
whether or not a particular stimulus has consequence of the correlation between the
its associative strength changed—that is stimuli and the reinforcement to which
'• of any substance.1 the subject is exposed during the course
This point has been represented in two of an experiment. In the language nor-
different ways. In theories such as those mally used by such theorists, the proba-
of Zeaman and House (1963) and Lovejoy bility of attending to relevant stimuli
(1968), the subject is assumed to attend typically increases, while the probability
to only one set of stimuli on a trial. The of attending to irrelevant stimuli typically
probability of attending to a particular decreases. In other words, if variations
stimulus on a particular trial, therefore, in stimulus A are correlated with changes
determines the probability that the out- in reinforcement, «A may increase, while
come of that trial will change the asso- if variations in stimulus B are not cor-
ciative strength of that stimulus. In the related with changes in reinforcement,
theory of Sutherland and Mackintosh «B may decrease.
(1971), on the other hand, subjects are The second major assumption common
assumed to switch in several analyzers on to all of these theories is that the proba-
a given trial. The strength of a particular bility of attending to one set of stimuli
analyzer, therefore, determines the amount is inversely related to the probability of
by which the associative strength of a attending to others. This is a necessary
particular stimulus is changed. Over a consequence of the "one-look" assumption
series of trials, of course, these two for- incorporated into the models of Zeaman
malizations amount to very much the same and House (1963) and Lovejoy (1968).
thing. More important, however, insofar Since subjects are assumed to attend to
as the strength of switching in a particular only one set of stimuli on a trial, it neces-
analyzer or the probability of attending sarily follows that as the probability of
to a particular set of stimuli simply de- attending to one set increases, so the
termines which stimuli change their asso- probability of attending to others must
ciative strengths (and by how much), this decrease. Moreover, even Sutherland and
characteristic of two-stage models is equally Mackintosh (1971), although allowing that
well represented by a single-stage model learning can occur about all available
that incorporates a learning rate parame- stimuli on a single trial, assume that the
ter, a, as exemplified in Equation 2. strengths of all analyzers sum to 1.0. In
Although theories of selective attention general, therefore, theories of selective at-
have, indeed, incorporated further assump- tention may be said to have assumed an
tions, these assumptions are also easily inverse relationship between the values
represented in a single-stage model as de- of a specific to the population of stimuli
terminants of the value of a. sampled in any given experimental situ-
The first of these assumptions, common ation. If that situation can be concep-
to all theories of selective attention, is tualized as containing only three sets of
that a is not a fixed consequence of such stimuli, A, B, and X, an increase in «A
physical characteristics of a stimulus as will be accompanied by corresponding de-
its intensity or modality, but rather that creases in either or both of «B and «x.
a may vary with the subject's experience. This assumption has been called the in-
The experience in question may have oc- verse hypothesis (Thomas, 1970).
curred before the start of an experiment, Although these two assumptions (that
the value of a rather than being fixed may
1
This is of course an oversimplification, for the change with experience, and that the
direction of the subject's attention is also assumed values of a. for different stimuli are in-
to affect which stimuli will actually determine versely related to each other) have both
performance at any given moment, but the ques-
tion of performance may for the moment be been incorporated into essentially all theo-
deferred. ries of selective attention, and although
both have also been rejected by the It was also Lawrence, of course, who
majority of other theories of associative first provided relatively convincing evi-
learning, it is important to see that they dence that changes in a might occur. His
are logically independent. There is, in experiments on simultaneous and succes-
fact, no reason why the possibility that sive discriminations (Lawrence, 1949, 1950)
subjects might learn to attend to relevant and on transfer along a continuum (Law-
stimuli or ignore irrelevant stimuli should rence, 1952) showed transfer from one
carry with it the implication that such discrimination problem to another, which
attention is selective in the sense required seemed to imply increases in the value
by the inverse hypothesis. One can con- of a for relevant stimuli and/or decreases
ceive of a theory of associative learning, in the value of a for irrelevant stimuli.
which while permitting a for various stim- This is not the place for a detailed review
uli to change in accordance with the of these studies, but although other ac-
schedule of reinforcement associated with counts of these transfer effects have been
those stimuli, does not insist that a change offered, there is reason to question their
in a to one set of stimuli is necessarily sufficiency (Sutherland & Mackintosh,
accompanied by an equal and opposite 1971). Similarly, although experiments
change in a to other sets of stimuli. Since on discrimination reversal undoubtedly in-
a theory that accepts one of the assump- volve a number of complex processes, it
tions common to most theories of attention can be argued that one reason why over-
without the other is thus clearly a logical training sometimes facilitates reversal
possibility, it is important that the pre- learning, and why exposure to a series of
suppositions and experimental evidence reversals of a single discrimination some-
relevant to the two assumptions be exam- times leads to eventually very rapid
ined independently. reversal, is that both overtraining and
serial reversal training increase the value
Changes in Attention with Experience of a for the relevant discriminative stimuli
Traces of the assumption that the direc- (Mackintosh, 1969; Mackintosh & Hoi-
gate, 1969). Finally, studies of intra-
tion of attention may change can be found
in both Lashley's (1929) and Krechevsky's dimensional and extradimensional shifts
have shown that the magnitude of transfer
(1932) early writings on discrimination
from one discrimination problem to another
learning. According to Krechevsky, sub- depends on whether the stimulus dimen-
jects entered the experimental situation
sions that were relevant and irrelevant in
with a particular set of hypotheses about the first problem retain their original sig-
which stimuli might be relevant and which nificance in the second problem (Shepp
irrelevant, but in the face of inconsistent & Eimas, 1964; Shepp & Schrier, 1969).
reinforcement would discard unsuccessful It is very difficult to see how such a
hypotheses and try out new ones. Al- finding can be explained without assuming
though hypotheses were thus said to be changes in a for relevant and/or irrelevant
"docile," the actual rules governing changes stimuli, which transfer to other stimuli
from one hypothesis to another were not falling along the same dimension.
specified, and the implication was that they
were better characterized as consequences It may be accepted that many of these
of a sudden new insight, or of the re- findings imply that a changes during the
organization of the subject's perceptual course of discrimination learning; they do
field, rather than being due to any readily not, however, enable one to specify the
understood process of associative learning. direction of such changes. As was im-
In this context, it was Lawrence (1949, plicitly acknowledged in the preceding
19SO) who first suggested that the "ac- paragraph, many of these results could as
quired distinctiveness of cues" might be well be a consequence of a decrease in a
a gradual learning process, amenable to for irrelevant stimuli, as of an increase
analysis in traditional terms. in a for relevant stimuli. It is somewhat
surprising, therefore, to find that formal by postulating changes in a stimulus-

theories of selective attention have con- specific learning-rate parameter such as a.
centrated on this second interpretation to Nor is it necessary to present a stimulus
the virtual exclusion of the first. It is in the complete absence of reinforcement,
true that Lawrence (1949) admitted that as in typical experiments on latent in-
a decline in the distinctiveness of irrelevant hibition, in order to retard subsequent
stimuli was probably at least as important conditioning. Indeed, sufficient exposure
as an increase in the distinctiveness of to uncorrelated presentations of a stimulus
relevant stimuli in generating the transfer and reinforcer may be an even more ef-
effects he observed. But in the formal fective procedure for reducing the asso-
two-stage models of discrimination learn- ciability of that stimulus with that rein-
ing proposed by Zeaman and House (1963), forcer (Mackintosh, 1973).
Lovejoy (1968), and Sutherland and Mac- Latent inhibition has also been reported
kintosh (1971), while differential reinforce- in experiments on free-operant instrumental
ment is assumed to increase attention to discrimination learning by Mellgren and
relevant stimuli, there is no comparable Ost (1969) and Halgren (1974). Although
independent mechanism for reducing at- a similar finding has not been reported in
tention to irrelevant stimuli. Except in discrete-trial studies, there is evidence of
the sense that a formerly relevant analyzer what would seem to be an analogous
will revert to its baseline strength if the effect. In experiments on simultaneous
stimuli it detects subsequently become and successive discrimination learning,
irrelevant, the only mechanism for de- Waller (1970) and Hall (Note 1) obtained
creasing the strength of irrelevant analyzers results that suggest that rats and pigeons
is by competition with relevant analyzers. may learn specifically to ignore irrelevant
It is, in fact, only the inverse hypothesis stimuli that vary from trial to trial without
that ensures that as attention to relevant being correlated with reinforcement.
stimuli increases, so attention to irrelevant There is, therefore, a considerable amount
stimuli must decrease. of evidence, consistent in a general way
There are good reasons to question this with the first assumption of theories of
analysis. A substantial body of evidence attention, that the associability of a stimu-
suggests that exposure to a set of stimuli lus with reinforcement is not an immutable
uncorrelated with reinforcement will spe- consequence of that stimulus's physical
cifically reduce the associability of those characteristics. Exposure to a set of
stimuli with subsequent changes in rein- stimuli correlated with changes in rein-
forcement, and that this is not just a con- forcement appears to increase the asso-
sequence of any increase in attention to ciability of those and similar stimuli with
another set of stimuli. The simplest dem- future changes in reinforcement. Con-
onstration of this effect is provided by versely, exposure to variations in a set of
experiments on latent inhibition (Lubow, stimuli in the absence of correlated changes
1973; Lubow & Moore, 1959), in which in reinforcement appears to decrease the
nonreinforced preexposure to a particular associability of those stimuli with future
stimulus interferes with subsequent con- changes in reinforcement. The fact that
ditioning to that stimulus. Experiments such changes in a occur, of course, does
by Rescorla (1971) and Reiss and Wagner not of itself specify the nature of the rules
(1972) have established that such non- for changing a that a formal model would
reinforced preexposure does not turn a have to incorporate in order to provide
stimulus into a conditioned inhibitor or a satisfactory account of the data. But
signal for nonreinforcement and indeed there is already a strong suggestion that
may significantly interfere with the estab- it is not enough to assume that attention
lishment of subsequent inhibitory condi- to irrelevant stimuli decreases simply by
tioning. It is difficult, therefore, to see virtue of an increase in attention to
how such results can be explained except relevant stimuli. It seems probable that
a symmetrical set of rules will be required the experimenter controls the subject's
that will permit opposite but independent exposure to the CS, the stimuli used are
changes in a to relevant and irrelevant typically salient, often unlocalized, and
stimuli. may sometimes last up to several minutes.
It stretches credulity to suppose that the
The Inverse Hypothesis rat's channel capacity is too limited to
The second main assumption held by permit processing of both a loud tone and
most theories of selective attention is that a bright light, when the two are switched
stimuli compete for attention, so that an on simultaneously for the duration of a
increase in attention to one set of stimuli 3-minute trial in an experiment on con-
will necessarily involve a decrease in at- ditioned suppression. Whether or not strict
tention to some other set of stimuli. It simultaneity of stimulus processing is im-
should be clear that the experiments re- possible, typical animal experiments have
viewed in the preceding section, designed imposed no such requirement.
to show that the probability of attending It is obvious, however, that the major
to a particular set of stimuli may change justification for a particular theoretical
with experience, do not have any bearing assumption is the extent to which it per-
on the validity of this second assumption. mits the derivation of experimental data.
What then has been the justification for What are the data that are typically
the inverse hypothesis? explained by the inverse hypothesis? They
One line of argument has been that it is may generally be described as instances of
a necessary corollary of the assumption, "stimulus selection," data which suggest
derivable from information theory, that that the association of one stimulus with
there are limits to any organism's capacity reinforcement does not depend simply on
to process information. Stimuli must be the characteristics of that stimulus, on its
assumed to compete for access to a chan- correlation with reinforcement, or even,
nel of limited capacity, and the formation apparently, on its past history, but may
of associations between one set of stimuli be affected by the nature of the other
and reinforcement will necessarily be at stimuli present at the same time, by their
the expense of the formation of associa- correlation with reinforcement, and by
tions with other stimuli. Plausible as it their past history. The most important
may seem to insist on limitations on the instances of such data are provided by
capacity to process information, however, studies of overshadowing and blocking.
it may be questioned whether these con- The term overshadowing refers to the find-
cepts are of any substantial relevance to ing, first reported by Pavlov (1927), that
the sorts of situations to which theories the presence of an equally relevant, more
of selective attention have been applied. salient stimulus may decrease or com-
Of the numerous differences between ex- pletely prevent conditioning to a less
periments on instrumental discrimination salient stimulus. A dog would associate
learning in animals, and those studies of a weak thermal stimulus with the delivery
dichotic listening or reaction times in of food, provided it was presented on its
human subjects on which the concept of own, but if it was presented only in con-
a limitation on capacity has been based, junction with a more intense auditory
perhaps the most important is that in the stimulus, no conditioning would accrue to
former situation subjects control their own the weaker stimulus. In general, condi-
exposure to the discriminative stimuli; tioning to B often proceeds more slowly
they are under no constraints to respond if B is always presented in conjunction
rapidly, and there is little reason to sup- with A than if B is presented alone.
pose, therefore, that they are processing The term blocking was used by Kamin
information at rates close to the limits (1969) to refer to the finding that prior
of their capacity. Although it is true that training on one element, A, of a com-
in experiments on classical conditioning pound stimulus, AB, might completely
prevent conditioning to the second ele- strengths of all stimuli (including A)

ment, B. If rats received a series of trials present at the moment of reinforcement.
on which a compounds CS, consisting of In an experiment on blocking, a sufficient
a light and a noise, signaled the occur- number of reinforced trials with A alone
rence of a shock, they would show sig- will ensure that FA approaches X ; at this
nificant conditioned suppression to both point, reinforced trials with AB will result
stimuli. But if these compound trials in little or no conditioning to B, since
were preceded by a series of trials on (X — F) will be nearly zero.
which, say, the light alone signaled shock, Rescorla and Wagner's model, although
they would show little or no suppression clearly quite distinct from any theory of
to the noise. selective attention, may still, without gross
Both overshadowing and blocking, it distortion, be said to explain stimulus
need hardly be pointed out, follow quite selection by appeal to an inverse hy-
directly from the inverse hypothesis. An pothesis. Stimuli are assumed to compete
intense stimulus is precisely one that is with one another, not for a limited atten-
associated with a high value of a, and its tional capacity but for a limited amount
addition to the stimulus compound cor- of associative strength conditionable by a
related with reinforcement will necessarily, given reinforcer. Other things being equal,
according to the inverse hypothesis, de- the greater the associative strength ac-
crease the value of a associated with any cruing to one element of a compound CS,
other stimulus. Similarly, prior training the less will be the associative strength of
with A relevant will increase <XA and hence the other. Unlike theories of selective
ensure that when B is added, «B will be attention, however, Rescorla and Wag-
low. In spite of this apparently happy ner's theory does not postulate what may
agreement between data and theory, how- seem to some implausible limits to an
ever, one can still ask whether blocking animal's capacity to attend to incoming
and overshadowing are explained more stimuli; moreover, as we shall see, there
satisfactorily by a theory of selective at- are cases where its predictions differ from
tention than by any other theory, and those derivable from theories of selective
whether they are observed on the occa- attention and cases where it is supported
sions required, and for the reasons speci- by the evidence. I shall argue, however,
fied, by such a theory. There is reason that in the final analysis, the data on
to believe that they are not. stimulus selection do not support any
Rescorla and Wagner (1972) proposed version of the inverse hypothesis: Blocking
an alternative explanation of these and and overshadowing are not the conse-
other instances of stimulus selection. Ka- quence of any simple competition between
min (1969) suggested that blocking occurs stimuli for some limited resource.
because on compound trials the subject
already expects the reinforcer and only EXPERIMENTAL ANALYSIS OF STIMULUS
unexpected reinforcers are reinforcing. SELECTION
Rescorla and Wagner have formalized one Blocking
aspect of this idea: The central assump-
tion of their model is that changes in the
If A is established as a signal for rein-
associative strength of a stimulus are in-
versely related, not just to the current forcement, and subjects are then exposed
to AB, signaling the same reinforcer, the
associative strength of that stimulus, but
available data suggest that the failure of
to the current strength of the entire stimu- conditioning to B is not due to the fact
lus complex of which it forms a part.
that A preempts attention but is rather
Formally,
a consequence of the fact that B signals
AF A = B(\ - F), (3) no change in reinforcement, predicting
nothing that is not already predicted by
where V is the sum of the associative the presence of A. When this condition
is not satisfied, blocking may not occur. would say that «B decreases whenever B
In Kamin's (1969) experiments on blocking signals no change from the prevailing or
in conditioned suppression, if A is estab- expected rate of reinforcement. The ex-
lished as a signal for a 1 mA shock, and pected rate of reinforcement, however, is
AB also signals a 1 mA shock, B acquires precisely that which is predicted by other
relatively little associative strength. But stimuli in the experimental situation.
if AB signals a stronger shock than that Thus the blocking experiment, in which
signaled by A alone, or no shock at all, AB signals no change in reinforcement
then significant excitatory or inhibitory from that predicted by A alone, is pre-
conditioning accrues to B. In the latter cisely the sort of situation that would
case, Wagner (1971) has shown that if result in a rapid decline in «B.
reinforced trials with A alone are followed It is, in fact, possible to show that such
by nonreinforced trials with AB, then an a decline in «B is one of the things that
increase in the number of reinforced trials happens when AB signals the same rein-
to A, so far from decreasing «B and the forcement previously signaled by A alone.
amount learned about B, actually ensures It will be recalled that if animals have
that B is established as an even stronger learned that A signals a 1 mA shock, the
conditioned inhibitor. presentation of AB, correlated either with
While it might still be possible in the a significant increase in the intensity of
face of this evidence to save the analysis shock or with no shock at all, results in
of blocking suggested by theories of se- significant excitatory or inhibitory condi-
lective attention, for example by assuming tioning to B. Mackintosh and Turner
that the values of a change when the (1971) found that a small number of
conditions of reinforcement change, it cer- trials on which AB signaled the same
tainly seems more appropriate to assume shock as A alone, interposed between
that any adequate explanation of blocking initial conditioning to A and subsequent
must take as its starting point the cardinal training with AB, significantly impaired
observation that blocking depends on the this excitatory or inhibitory conditioning.
redundancy of the added element: Block- The implication is that <*B decreased be-
ing occurs because AB signals the same cause when B was first introduced it
reinforcer as that signaled by A alone. signaled no change in the probability or
This is, of course, precisely the position magnitude of reinforcement.
adopted by Kamin (1969) and Rescorla It is, of course, one thing to say that
and Wagner (1972). But there is another if AB signals the same reinforcement pre-
way of explaining blocking in terms of viously signaled by A alone, «B will de-
redundancy, one which appeals neither to crease. It is another thing to say that
limitations of attentional capacity nor to this is the sole or even a major cause of
limitations on the total associative strength blocking. This assertion has a number of
conditionable by particular reinforcers. further implications. If the blocking of
It is possible that the redundancy of B conditioning to B depends on subjects
reduces conditioning to B because it learning to ignore B, for example, no
reduces as. blocking can be expected until they have
I have already argued that any adequate done so. Where blocking is substantial,
theory of associative learning must allow it must be assumed that changes in asso-
a to change and that this change must ciative strength are slow relative to
include a rule for decreasing a to stimuli changes in a. If conditioning proceeded
uncorrelated with changes in reinforce- more rapidly, one might expect to see
ment. If it were assumed that «A in- relatively little blocking. But even under
creases whenever A signals a change from normal circumstances, one would expect
the prevailing or expected conditions of to see some conditioning to B, since it
reinforcement, the symmetrical rule for must take some minimal number of trials
decreasing attention to irrelevant stimuli (at least one) to reduce C*B to a point
where further changes in the associative was, of course, precisely this feature that
strength of B are minimal. made it possible to explain blocking in
There is, in fact, some evidence con- terms of a reduction in a, rather than in
sistent with this final implication. Kamin terms of the inverse hypothesis. But in
(1969) found it possible to detect some experiments on overshadowing, animals
conditioning to B in his standard experi- may be trained with AB-signaling rein-
mental situation; more important, his forcement from the outset. If A has not
data suggested that most if not all of this already been established as a signal for
conditioning was a consequence of the reinforcement, how can its presence de-
first reinforced AB trial. This observa- tract from conditioning to B during rein-
tion agrees with some results of an experi- forced compound trials, unless some appeal
ment on blocking in instrumental learning is made to the idea that stimuli compete
by Turner (Note 2). In this experiment, with each other for association with rein-
although pretraining on A resulted in some forcement? If a subject simply receives
blocking of learning about B during AB reinforced trials with an AB compound,
trials, blocking was far from complete. both A and B should be established as
Since animals receiving 100 AB trials signals for reinforcement, and without in-
learned no more about B than those re- voking the inverse hypothesis it is hard
ceiving only 10 AB trials, however, the to see why the presence of one should
implication is that learning about B was cause any decline in attention to the other.
confined to the first few trials after its There are, however, at least two circum-
introduction. stances that would appear to provide a
Although these results are consistent sufficiently close parallel to the conditions
with the present argument, they do not of an experiment on blocking to make it
bear on the strong prediction derivable possible to apply the principle of learned
from this analysis, that little or no block- irrelevance to the case of overshadowing.
ing should occur on at least the first If A is better correlated with reinforcement
reinforced compound trial. A satisfactory than B, or if A is a much more salient
test of this prediction requires a situation stimulus than B and therefore conditions
in which reliable conditioning occurs to more rapidly, then there is a sense in
the elements of a compound CS after a which B will be a redundant signal of
single reinforced trial. The conditioned reinforcement. In the former case, A, by
suppression of licking in thirsty rats by virtue of its superior schedule of rein-
a CS signaling shock provides one such forcement, will acquire associative strength
situation, and we have attempted to see more rapidly than B. In the latter case,
whether under these circumstances pre- even though the schedule of reinforcement
training on A does result in significant associated with A and B is identical, since
blocking of conditioning to B when only A conditions more rapidly than B, A will
one reinforced AB trial is given (Mackin- be established as a reliable signal for rein-
tosh, Note 3). In several experiments, no forcement after a number of reinforced
evidence of blocking could be found. compound trials, while B will have ac-
These results suggest, therefore, that a quired little associative strength. At this
substantial part of blocking observed in point, just as in the case where the schedule
other situations may indeed be a con- of reinforcement associated with A is better
sequence of a rapid decline in attention than that associated with B, the situation
to the added, redundant element. will be analogous to that obtaining in
experiments on blocking: The presence of
Overshadowing B predicts nothing that is not already
The design of experiments on blocking predicted by A alone, and aB will decline.
is such that the added stimulus B predicts Numerous studies of compound condi-
no change in the level of reinforcement tioning have established that of two
from that predicted by A alone, and it equally salient stimuli, the more valid
stimulus, associated with a better schedule intense auditory and visual stimuli in
of reinforcement, will overshadow the less another study of conditioned suppression,
valid (e.g., Egger & Miller, 1963; Wagner, found no evidence of overshadowing at all.
1969; Wagner, Logan, Haberlandt, & It is true that in some of these experi-
Price, 1968) and that of two equally valid ments ceiling effects may have obscured
stimuli, the more salient will overshadow any evidence of overshadowing. In order
the less salient (e.g., Kamin, 1969; Miles to obtain further evidence on this point,
& Jenkins, 1973; Pavlov, 1927). What we have studied overshadowing in condi-
has not been recognized, however, is that tioned suppression, using noise and light
these appear to be the only reliably as CSs, and varying the intensity of the
established instances of overshadowing. noise between 50 and 85 db. re 20 /iN/m 2 .
Although the more salient element of a To counteract ceiling effects, subjects
compound CS overshadows the less salient, were tested over a series of trials in ex-
there is no evidence that the less salient tinction, to the point where no group
will detract from conditioning to the more maintained complete suppression. Al-
salient, nor is there much evidence to though there was some evidence of re-
suggest that if two equally salient stimuli ciprocal overshadowing at intermediate
are associated with the same schedule of intensities of the noise, in the extreme
reinforcement, one will overshadow the cases there was no evidence of any such
other. interaction. The most intense noise sig-
If stimuli compete for association with nificantly overshadowed the light, while
reinforcement in the manner suggested by the least intense noise was itself signifi-
the inverse hypothesis, one would expect cantly overshadowed by the light. In
to observe a reciprocal interaction between neither instance, however, did the over-
all elements of a compound CS. If ani- shadowed element, even though itself ac-
mals receive a series of reinforced trials quiring significant associative strength,
with AB, then, other things being equal, detract from conditioning to the stronger
any conditioning to B must always be at element.
the expense of conditioning to A, and In experiments on instrumental dis-
vice versa. The experimental evidence crimination learning, significant overshad-
does not support this expectation. Pavlov owing has been observed, provided that
(1927) reported that overshadowing was the overshadowed cue is either extremely
a matter of a reduction in conditioning to difficult to discriminate (Lovejoy & Rus-
the weaker of two elements of a com- sell, 1967) or both relatively difficult and
pound CS as a consequence of the presence in a different location from the more
of the stronger. The unspoken implica- salient, overshadowing cue (Sutherland &
tion was that conditioning to the stronger Andelman, 1967). When neither of these
was unaffected. Although Pavlov did not conditions is satisfied, overshadowing is
present the data to support this inference, not observed (Sutherland & Andelman,
the conclusion is entirely consistent with 1967; Turner, 1968), and it may be dif-
subsequent experiments. ficult to detect any evidence of an inter-
Kamin (1969) and Mackintosh (1971), action between the stimuli (Sutherland &
in studies of conditioned suppression, re- Holgate, 1966; Warren, Derdzinski, Hi-
ported that a bright light would over- rayoshi, & Mumma, 1970). The most
shadow auditory stimuli of varying inten- extensive study of overshadowing in in-
sities, but neither found any evidence to strumental discrimination learning is that
suggest that the presence of the auditory by Miles and Jenkins (1973). Different
stimulus detracted from conditioning to groups of pigeons were trained on a suc-
the light, nor that the amount of condi- cessive discrimination with either the
tioning to the light was an inverse func- presence or absence of a tone, a dif-
tion of the intensity of the auditory ference in light intensity, or a combination
stimulus. Schnur (1971), using relatively of both cues serving to distinguish posi-
tive and negative trials. The discrimi- Stimuli may condition at different rates,
nability of the light cue varied across and at least one cause of these differences
groups. The most striking feature of their may be represented, as in Equation 2, by
results is that the presence of the tone differences in a stimulus-specific learning-
detracted from control by the light only rate parameter, a. The value of a is
with the least discriminable value of the initially determined by the physical char-
light, but for this group the presence of acteristics of the stimulus and the subject's
the light did not significantly detract from sensory apparatus, but it may also change
control by the tone. Conversely, the more with experience. In particular, if stimu-
discriminable lights did overshadow the lus A is correlated with changes in rein-
tone, but in these groups the presence of forcement, «A will increase, and if stimulus
the tone did not detract from control by B is not correlated with changes in rein-
the light. No group, in other words, forcement, «B will decrease. The critical
showed simultaneous overshadowing of departure from traditional theories of se-
both light by tone and tone by light. lective attention is that in a situation
It is too early to reach any unequivocal where several stimuli, A, B, C, are pre-
conclusion, but the weight of the evidence sented, an increase in a A does not neces-
does not appear to provide much support sarily or directly cause a decline in «B
for the inverse hypothesis. Overshadowing or ac- Changes in CXB and «c are inde-
does not appear to be the general, recip- pendent of changes in a&. They are caused
rocal affair required by any theory that solely by the correlation of B and C with
assumes stimuli necessarily compete for a changes in reinforcement and do not de-
limited total amount of associative strength pend in any direct way on changes in a&.
or for access to a limited-capacity channel.
To a first approximation, stimuli that are Rules for Changing a
overshadowed appear to be only those Even if it is assumed that there may be
with lower salience or inferior correlation
changes in attention to particular stimuli,
with reinforcement, which results in a the assumption that such changes in at-
slower rate of conditioning than that ac- tention are entirely independent at first
cruing to concurrently presented, over- sight seems to rule out the possibility of
shadowing stimuli. Any interaction be- explaining the types of interaction between
tween elements of a compound that is stimuli that are exemplified by the phe-
confined to this sort of situation may be
nomena of blocking and overshadowing.
quite satisfactorily explained by a theory This is the conclusion accepted by Fisher
that postulates reductions in a to stimuli and Zeaman (1973), who have recently
that predict no unexpected changes in proposed a modification of the original
reinforcement. Zeaman and House (1963) model, one of
SPECIFICATION OF A NEW THEORY the new features of which is a rejection
OF ATTENTION of the inverse hypothesis. They noted
that their revised theory would not pre-
If the data traditionally thought to dict blocking or overshadowing under
support one of the assumptions adopted normal conditions. As the arguments of
by theories of selective attention can be the preceding sections perhaps have shown,
explained in other ways, it may be time the validity of this conclusion depends on
to drop that assumption. The theory that the nature of the rules proposed for
emerges from the considerations advanced changing attention. Fisher and Zeaman
so far may thus be characterized as follows. (1973) follow Sutherland and Mackintosh
Changes in the associative strength of a (1971) in assuming that the strength of a
stimulus do not depend only on its cor- particular observing response will increase
relation with reinforcement or on the whenever the outcome of a trial confirms
magnitude of that reinforcement; they the subject's expectations about the stimu-
also depend on the nature of the stimulus. lus detected by that observing response;
similarly, disconfirmation of an expecta- interpretation of stimulus selection. If

tion results in a decrease in the strength overshadowing is not the general and
of that observing response. Rules such reciprocal phenomenon postulated by the
as these will not, of course, predict any inverse hypothesis, and if little or no
decline in the strength of an observing blocking occurs on the first trial on which
response to the redundant stimulus in an a new component is added, this is evidence
experiment on blocking. If a subject re- against Rescorla and Wagner's analysis
ceives a series of reinforcements signaled just as much as against the version of the
by AB, the fact that these trials were inverse hypothesis postulated by theories
preceded by a series of reinforced trials of selective attention. The force of the
signaled by A alone will have no effect previous argument was precisely that
on any changes in the probability of overshadowing and blocking are conse-
attending to B. The rules for changing quences of reductions in the value of a
observing responses make no reference to associated with overshadowed and blocked
whether a particular reinforcer is or is not stimuli and are therefore to be explained
expected on the basis of other simulta- by the postulation of appropriate rules for
neously presented stimuli. changes in a.
A theory such as Fisher and Zeaman's, It remains, then, to propose such rules.
therefore, in which observing responses Instead of saying, as do Sutherland and
change independently and simply in ac- Mackintosh (1971) and Fisher and Zeaman
cordance with the reinforcement schedule (1973), that subjects learn to attend to
associated with each independent set of and ignore stimuli to the extent that those
stimuli, while able to explain data on the stimuli successfully predict the outcome
acquired distinctiveness of cues, reversal of a trial, we want to say that it depends
learning, and intradimensional and extra- on whether the stimuli are uniquely suc-
dimensional shifts, will not predict such cessful in their predictions. The intuition
phenomena of stimulus selection as block- that we require to formalize is that «A
ing and overshadowing. It is possible, of should increase if A predicts an otherwise
course, that such phenomena are in no unexpected reinforcer, while «A should de-
sense attentional in origin. One might crease if A signals no change in reinforce-
argue that the phenomena of acquired ment from the level expected on the basis
distinctiveness and stimulus selection could of other events. There are presumably,
(and should) be handled by entirely dif- a number of ways in which this might be
ferent theoretical constructs. An adequate done, but possibly the simplest is as fol-
theory of associative learning would be one lows. The extent to which a reinforcer
that explained acquired distinctiveness by is predicted by A is represented by the
allowing a to change in the manner pro- absolute value of the term |X — FA |,
posed by Fisher and Zeaman (1973), and where, as usual, X is the asymptotic asso-
that explained stimulus selection in the ciative strength conditionable by that re-
manner proposed by Kamin (1969) and inforcer, and FA is the current associative
Rescorla and Wagner (1972). strength of A. If we wish a.^ to increase
Although a case could be made for such whenever the outcome of a trial is pre-
a hybrid theory, it must be recognized dicted by A better than by all other
that Rescorla and Wagner's analysis of events on that trial, we could say,
blocking and overshadowing is a special
AaA is positive if | X — FA < | X — Fx |,
case of the inverse hypothesis, since in-
creases in the associative strength of one (4)
component of a compound CS are assumed where Fx is the associative strength of all
to be at the expense of other components. stimuli other than A present on that trial.
As we saw earlier, therefore, the analysis Conversely, if we wish <XA. to decrease
is vulnerable to some of the arguments whenever the outcome of trial is predicted
that can be advanced against any such by other events at least as well as by A,
we could say, Overshadowing will also be predicted if

A is a better predictor of reinforcement
A«A is negative if | X — FA | > X — Fx •
than B, for example if reinforced trials
(5) signaled by AB are alternated with non-
Stimuli can, of course, be established as reinforced trials signaled by B alone.
signals for nonreinforcement just as for Under these conditions, the different sched-
reinforcement. The simple assumption ules of reinforcement associated with the
that the value of X for nonreinforcement two stimuli will result in faster condi-
is either zero or some negative number tioning to A than to B, and the consequent
will permit appropriate changes in a on reduction in «B will interfere with further
nonreinforced trials.2 conditioning to B. Similar arguments
It is easy to see how such rules will would apply to the other examples of
predict the occurrence of overshadowing relative validity studied by Wagner (1969).
and blocking under appropriate circum- There is, however, one serious problem
stances; «B will decrease if B is always with this analysis. There is evidence that
presented in conjunction with A and A has a less valid stimulus will not only fail to
previously been established as a signal for gain associative strength but also may
the reinforcer in question, or if, by virtue apparently lose such strength as it had
of its greater salience, A is initially asso- acquired during an earlier history of rein-
ciated with a higher value of a and thus forcement. Wagner et al. (1968) showed
acquires associative strength faster than B. that a light accompanied by a pair of
If it is further assumed that the size of tones would acquire significant control
the change in a is proportional to the over responding, but that when the cor-
discrepancy between | X — FB | and relation of the tones with reinforcement
X — Fx , then «B will decrease most was increased, the light lost control, even
rapidly in the blocking situation, where though its own correlation with reinforce-
on the first trial on which B is introduced, ment was uncharged. Similarly, if B is
FB is near zero and Fx (which includes FA) more salient than A, it may initially ac-
is near X. Conversely, «A will increase quire control, even though it is a less
faster if reinforcement is signaled by A valid predictor of reinforcement. As train-
alone than by a compound CS that in- ing continues, however, the difference in
cludes other stimuli as well as A. In validity outweighs the difference in sali-
principle, therefore, some reciprocal over- ence, and B loses control while A gains
shadowing would be predicted, but not if control (e.g., Jenkins, 1973; Rescorla,
the stimuli were sufficiently salient to be 1972b). An analysis that accounts for
associated with high values of a at the overshadowing by saying that a less valid
outset of conditioning. The data on the stimulus fails to acquire associative
effects of salience on overshadowing pres- strength because of a decrease in a is in
ently available are not sufficient to say no position to explain how it can lose
whether such an assumption is justified; strength that it has earlier acquired.
it might prove necessary to adopt rather The simplest explanation of such results,
different rules in light of further research. consistent with the present theory, would
2
This should not be construed as unconditional be to assume that a affects both learning
acceptance of Rescorla and Wagner's (1972) as- and performance. Even if its associative
sumption that reinforcement and nonreinforcement strength remained high, then, a stimulus
both result in changes in a single variable, V. might lose control over responding if a
For present purposes, one could equally well as-
sume that the omission of an expected reinforcer decline in a. decreased the probability that
resulted in an increment in some specific inhibitory this associative strength would be trans-
process, which grew to some asymptote, X/, and lated into performance. If B acquires
that the net associative strength, V, of a stimulus control over responding, but is then pre-
was determined by subtracting this inhibitory
process from a separate excitatory process which sented only in conjunction with a second
was incremented on reinforced trials. stimulus (A) better correlated with rein-
forcemeat, any decline in control by B may occur whenever the added element
may reflect the eventual decline in aB predicts some event of consequence (such
brought about by the relative change in as the second shock in Kamin's experi-
B's validity. It is too early to say whether ment), whether or not that event is itself
this is a satisfactory solution; as noted able to support conditioning. According
later, this general case provides the only to the present analysis, one could argue
compelling reason for assuming that a af- that the function of the surprising event
fects performance as well as learning. was to prevent a decline in the value of
The present theory states that blocking a associated with the added element, thus
depends on a decline in a to the added enabling conditioning between that element
element; as such, blocking should never and the original unconditioned stimulus
be complete and will depend on the rate (UCS) to proceed normally.
of change in a. The most obvious factor The application of the present rules to
determining the degree of blocking will be the case of acquired distinctiveness and
the extent to which the added stimulus dimensional transfer in instrumental dis-
signals some change in reinforcement. crimination learning is relatively simple.
If A alone signals reinforcement and AB Relevant stimuli in a discrimination prob-
signals nonreinforcement, then aB will tend lem predict the occurrence and omission
to increase rather than decrease, since B of reinforcement more accurately than any
must be a better predictor of nonrein- other stimuli, and an increase in their a
forcement than A. If AB signals a stronger values will follow from Equation 4. Simi-
reinforcer than that signaled by A alone, larly, irrelevant stimuli, uncorrelated with
then «B will tend to decrease, but less the delivery of reinforcement, are presented
rapidly than it would if AB signaled the in conjunction with predictive relevant
same reinforcer as A. This should pre- stimuli and should suffer a decline in
sumably be sufficient to predict the un- a value.
blocking observed by Kamin (1969) in his The prediction of the effects of non-
experiment in varying shock intensity, reinforced preexposure to a CS and of un-
especially when it is remembered that in correlated presentations of a CS and UCS
his experiment AB signaled a 4 mA shock, requires that such a CS be regarded as
and this would have been sufficient to forming a compound with a set of back-
produce substantial levels of conditioning ground stimuli, C. The crucial point, then,
to B in one or two trials. There is, how- is that the presentation of AC (CS plus
ever, reason to believe that unblocking background) predicts no change from the
may be produced by the occurrence of probability of reinforcement or nonrein-
any surprising event shortly after the forcement predicted by C alone, and a de-
presentation of the compound stimulus. cline in CCA should follow. There is, how-
Kamin (1969) reported that the delivery ever, some question whether this is an
of a second shock 5 sec after each com- ideal account. In the first place, it is only
pound trial resulted in significant condi- the phenomenon of latent inhibition that
tioning to the added element, even though necessitates the otherwise rather unhappy
this double shock did not appear to act assumption that <*A declines even when
as a stronger reinforcer. Gray and Ap- I X — VA | is equal to | X — Fx |. For
pignanesi (1973) observed a similar effect reasons of theoretical symmetry, if for
when they presented a brief auditory and no others, one would expect this equality
visual stimulus shortly after each com- to produce no change in a&. In the second
pound trial. If confirmed, such results place, there is reason to believe that
would suggest that unblocking was not random presentations of a CS and UCS
simply due to an increase in X enabling retard subsequent conditioning more se-
the added element to acquire associative riously than does simple nonreinforced
strength, in the manner suggested by preexposure to the CS; this additional
Rescorla and Wagner (1972). Unblocking effect, moreover, may be relatively specific
to conditioning with that particular UCS Starting Value of a: Representation of

(Mackintosh, 1973). One interpretation of Stimulus Salience
this observation is that a is both stimulus-
and reinforcer-specific. Changes in the It is not enough to devise rules for
associability of a tone and shock, result- changing a. A comprehensive theory must
ing from exposure to particular correlations specify the limits within which a may
between the tone and shock, will produce change and the conditions determining
only generalized changes in the associ- the starting value of a. The simplest
ability of that tone with other reinforcers. solutions to these problems are to allow a,
If further research were to confirm this as a learning-rate parameter, to vary be-
suggestion, the notion of a as a stimulus- tween 0 and 1 and to assume that the
specific learning-rate parameter would need starting value of a. for any particular
replacing with a parameter «A,R, the value stimulus is positively correlated with the
of which determined the magnitude of intensity or salience of that stimulus.
changes in the associative strength of A While these are reasonable assumptions,
when it was paired with reinforcer R (or they are not in fact sufficient to account
the omission of that reinforcer). If a thus for the effects of stimulus salience.
represented the associability of a particu- Lovejoy (1968), in his formalization of
lar stimulus with a particular reinforcer, a two-stage theory of selective attention,
changes in a would be produced only by distinguished between the fixed, base-level
exposure to that stimulus and reinforcer distinctiveness of a cue and its changeable
(or similar ones), and the phenomenon of or directable distinctiveness. Only the
of latent inhibition would lie outside the latter was subject to modification by
scope of the theory—to be explained, learning. The implication was that changes
perhaps, in terms of some simpler mecha- in attention to various cues due to ex-
nism of habituation. perience are "bounded, second-order effects
In the absence of further evidence, which take place within definite limits
further speculation along these lines would that depend on the initial strength of the
be idle. For the present, the rules sug- cues" (p. 60). The type of consideration
gested in Equations 4 and 5 may be that led Lovejoy to this solution was the
regarded as one way of representing the observation that however much one might
idea that the associability of a stimulus pretrain rats to attend to brightness cues
with reinforcement will be affected by the in a jumping stand, they are unlikely to
predictiveness of that stimulus relative to learn a simultaneous brightness discrimi-
that of other, concurrently presented nation as rapidly as a spatial discrimina-
stimuli. These rules are certainly not to tion. If the difference between spatial and
be thought of as final, but rather as brightness discriminations was represented
illustrating that it is possible to express only by differences in the starting value
this informal idea reasonably precisely. of a parameter such as a, and if sufficient
These particular rules, it will be recog- exposure to a correlation between a set
nized, bear a more than passing resem- of stimuli and reinforcement resulted in
blance to Rescorla and Wagner's (1972) the a value for those stimuli tending to
rules for changing V: In their system, 1.0, it would follow that differences in the
a change in FA depends on the status of difficulty of visual and spatial discrimina-
both FA and Fxl here a change in «A tions would eventually disappear. It seems
depends on the relative status of FA clear that differences in the difficulty of
and Fx. Informally, they assume that instrumental discriminations cannot be
only if a reinforcer is otherwise unpredicted adequately represented in this simple way;
will its presentation in conjunction with A animals may never perform with complete
increase the associative strength of A; here accuracy if the discrimination is hard
it is assumed that only if a reinforcer is enough (e.g., Hara & Warren, 1961), and
otherwise unpredicted will its presentation Mackintosh (1969), for example, noted
increase the probability of attending to A. that differences in the speed of reversing
an easy or difficult brightness discrimina- S— that will result from each reinforce-
tion after extensive overtraining were not ment of S-f, and the greater the general-
predicted by such a model. ization of inhibition to S+ that will result
It is equally easy to show that such an from each nonreinforcement of S— (cf.
account is inadequate for the data of clas- Kendler, 1971).
sical conditioning. Although the main These ideas can be formally represented
effect of differences in the intensity of the in a relatively simple way by introducing
CS is a difference in the rate of condi- a parameter, 5,-,y, representing the simi-
tioning, Kamin (1965) has shown that a larity of the itii and jth stimuli, such that
very weak CS may result in a lower level 0 < S < 1. With two discriminative stim-
of conditioning even at asymptote. uli, Ai and A 2 , reinforcement of Ai will
Asymptotic differences in performance, of result in an increment in the associative
course, cannot be attributed to variations strength of AI according to the usual
in a learning-rate parameter. equation:
Differences in the discriminability of
S+ and S— in a discrimination problem AF Al = « Al <KX - FAl). (2)
or between the presence and absence of By generalization from AI to A2, however,
the CS in classical conditioning, therefore, this trial will also result in an increment
seem to have more permanent effects on in the associative strength of A2, according
performance than can be encompassed by to the equation,
a model that represents such differences
by assigning different value's to a, but A FA = - FAl). (6)
which then permits a for any stimulus to Similarly, nonreinforcement of A2 will de-
vary between 0 and 1 as a result of ex- crease the associative strength of A2, and
perience. Instead of introducing a further will also result in a proportional decrease
parameter (such as a fixed a') for each in the associative strength of AI.
set of stimuli, to represent their dis- A sufficiently high value of SAI.AJ will
criminability from the background or from ensure that discriminative performance be-
other discriminative stimuli, it would seem tween AI and A2 will never be perfect,
more consistent with other psychological just as a sufficiently high value of SA,A
knowledge to rely on the notion of gen- (strong generalization between A and its
eralization. Perkins (1953) and Logan absence) will ensure that even at asymp-
(1954), for example, argued that the well- tote the level of conditioning maintained
authenticated observation that condition- by a weak CS will be below that main-
ing proceeds more rapidly with an intense tained by a more intense CS. With
CS than with one less intense may be a suitable performance rules, intermediate
consequence of differences in generalization levels of generalization will not affect per-
between the CS and the background. The formance at asymptote, but the general-
more intense a CS, the greater the dif- ization parameter will still provide an
ference between that CS and its absence additional factor (over and above a) de-
(the background). Reinforcement in the termining rate of learning. It is possible,
presence of an intense CS and nonrein- therefore, to allow that with sufficient
forcement in its absence, therefore, will training the value of a associated with
result in less generalization of excitation any stimulus can approach 1, without
to the background and less generalization thereby abolishing differences in rate of
of inhibition from the background to the learning correlated with differences in in-
CS than will a comparable amount of tensity or discriminability.
differential reinforcement correlated with
the presence and absence of a less in-
Specificity of a
tense CS. By the same line of argument,
the greater the difference between S+ Most formal theories of selective atten-
and S— in a discrimination problem, the tion have followed Zeaman and House
greater the generalization of excitation to (1963) and Sutherland (1964) in assuming
that subjects learn to attend not just to analyzers appropriate to entire stimulus
specific stimuli, such as a vertical line, dimensions are strenghtened and weakened.
black door, or red response key, but to One could equally well assume that an
stimulus dimensions, such as line orienta- increase in attention to one stimulus gen-
tion, brightness, or hue. Sutherland and eralized to other stimuli in proportion to
Mackintosh (1971), for example, assumed their similarity to the training stimulus.
that changes in attention could be repre- Thus an increase in a^v resulting from
sented by changes in the strengths of the correlation of reinforcement with the
various analyzers, each of which detected presence of AI and nonreinforcement with
variations along a particular stimulus di- its absence, might result in graded in-
mension. The assumption is a rather creases in a\v OAS . • • «An- Similarly, if
natural one for theories designed to ac- AI signaled reinforcement and A 4 signaled
count for the data of experiments on nonreinforcement, as in an experiment
simultaneous discrimination learning, where on discrimination learning, generalized
subjects are usually confronted on each changes in a would accrue to the inter-
trial with a pair of stimuli, such as vertical vening stimuli A2 and AS from both AI
and horizontal lines or black and white and A<I.
doors, which can clearly be described as This approach has the advantage of
differing along such dimensions. As Jen- again narrowing the discrepancy between
kins and Sainsbury (1969) have noted, traditional, single-stage theories of asso-
however, it may not seem quite so natu- ciative learning and two-stage theories of
rally suggested by a successive discrimina- selective attention. No appeal is made
tion between the presence and absence of to hypothetical observing responses or
a specific stimulus, and it can be applied analyzers detecting dimensions of stimulus
only with some difficulty and little success change. A change in the state of one
to certain of the data obtained in such stimulus, whether of its associative
a situation. strength, V, or of its learning rate param-
There are, of course, experimental data eter, a, is said to generalize to other
that seem to provide good evidence for the stimuli in accordance with entirely tradi-
dimensionality of attention. In studies of tional (even if unspecified) assumptions.
intradimensional and extradimensional There is the further advantage that this
shifts, the specific stimuli to which the revision may explain the relatively small
subject is exposed change from one stage differences often observed in comparisons
of the experiment to another; the observa- of intradimensional and extradimensional
tion that learning of the shift problem is shifts in animal subjects. If changes in a
faster if the relevant stimuli of the first to one pair of stimuli differing along a
problem differed along the same dimension particular dimension result in only gen-
as those of the shift certainly suggests that eralized changes in a to another pair, it is
the changes in' attention resulting from entirely reasonable to expect these latter
original training were changes in attention changes to be rather small.
to entire stimulus dimensions. Similarly,
if the phenomenon of transfer along a con- Performance Rules
tinuum (Lawrence, 1952) is to be inter- Theories of learning are inferences from
preted in attentional terms, the implica- observed behavior; some relationship must
tion is that an increase in the probability be postulated between the theorist's terms
of attending to stimuli differing widely and the experimental subject's responses.
along a particular dimension results in an It will be sufficient, for present purposes,
increase in the probability of attending to to suppose that the value of FA bears
another set of stimuli differing less widely some monotonic relationship to the proba-
along the same dimension. bility that A, as a CS, will elicit an ap-
It is questionable, however, whether data propriate CR, or as a discriminative
such as these require the assumption that stimulus, will result in the occurrence of
an appropriate instrumental response. Al- one position over a long series of trials,
though deliberately vague, this is neither may make no errors once they break their
contentious nor problematic. Theories of position habit (Sutherland & Mackintosh,
attention, however, are faced with the 1971, p. 91). Analysis of the performance
necessity of specifying further relations of animals shifted from one discrimination
between their theoretical terms and their to a second, in which the originally relevant
subjects' responses. As noted earlier, in stimuli become irrelevant, has shown that
traditional two-stage theories the direc- few if any errors occur on those trials
tion of the subject's attention is assumed when the new S-f- is combined with the
to affect not only what the subject learns former, but now irrelevant, S+. Such
as a consequence of the outcome of the results suggest that both stimuli contribute
trial, but also what response the subject to the subjects' choice behavior (Medin,
makes on that trial. The probability that 1973; Tighe, 1973).
a rat will choose the black rather than These conclusions are quite consistent
the white arm of a T maze depends not with the general trend of the present
only on the relative associative strengths argument. It may be assumed, therefore,
of black and white, but also on the rela- that all stimuli contribute to performance
tive strengths of attention to brightness in accordance with their current associative
and other stimuli. In the original one- strengths. The question remains whether
look model of Zeaman and House (1963), one should assume that changes in a not
the subject attends to only one stimulus only determine the rate of learning about
dimension on each trial, and behavior on specific stimuli but also affect the extent
that trial is controlled only by the response to which a stimulus' associative strength
tendencies to those stimuli. Sutherland is actually translated into performance.
and Mackintosh (1971) adopted a more One could certainly assume that the con-
complex solution, assuming that behavior tribution of any stimulus to performance
was sometimes controlled by only one was a product of its associative strength
analyzer, but that sometimes more than and a value. Thus as UA. declined, even
one set of stimuli would determine if FA had some moderate value, the con-
behavior. tribution of A to performance would also
There is, in fact, definitive evidence decline.
that in two-choice discriminations behavior There is actually very little evidence
can be controlled simultaneously by several bearing on this question. Since, in gen-
sets of stimuli; rats can, after all, solve eral, a stimulus with a low value of a. will
conditional discriminations. But there is also be one with little associative strength,
equally good evidence pointing to this it may be very difficult to decide whether
conclusion from studies of simple simul- the reason why a particular stimulus fails
taneous discriminations, where solution to affect behavior is because it is not
could in principle be based on only one attended to (i.e., has a low a value) or
set of stimuli at a time. When more than because its associative strength is low.
one set of stimuli is simultaneously (but There is evidence from studies of dis-
redundantly) relevant, it is possible to crimination learning that differences in the
show that performance may be controlled associative strength of S+ and S— from
by both on a single trial (Sutherland & an initial discrimination are preserved
Mackintosh, 1971, p. 142). Even in the while the animal learns another problem
case where one set of stimuli is relevant with the initial stimuli no longer corre-
and another irrelevant, there is evidence lated with reinforcement—even though
that both relevant and irrelevant stimuli there may be no detectable evidence that
may combine to control performance. such stimuli affect behavior (e.g., latency
This is suggested by the fact that rats, of responding) on the intervening problem
trained on a simultaneous visual discrimi- (Mackintosh, 1963; Stettner, 1965). Simi-
nation, having responded consistently to larly, experiments on generalization in
pigeons have established that in the CONCLUSIONS

presence of one stimulus (e.g., a colored The theoretical ideas proposed here may
background on the response key), another be summarized briefly. Changes in the
set of stimuli, such as the orientation of associative strength of a stimulus are
a line on the key, may exert little or no partly determined by a learning-rate pa-
control over responding, in spite of the rameter, a, specific to that stimulus. This
fact that a particular line has been estab- parameter is itself subject to change, in-
lished as a signal for reinforcement, and creasing when a stimulus predicts a change
reliably sloping gradients may be observed in reinforcement, decreasing when it does
if different lines are presented on a black not. This idea is formally equivalent to
background (Newman & Benefield, 1968). one of the main tenets of two-stage, at-
Such instances of masking of control, tentional theories of learning, namely, the
however, can be explained without recourse assumption that the probability of at-
to the assumption of competition between tending to a stimulus determines the
stimuli for control. The associative strength probability of learning about that stimulus
of the masking stimuli may be sufficiently and may itself change with experience.
great that ceiling effects obscure the con- This equivalence may justify character-
trol being exercised by the masked stimuli. izing the present set of ideas as a theory
These results, therefore, cannot really be of attention, but since that term has a
said to provide conclusive evidence of the number of connotations, it might be better
influence of a on control. Until such to stress that what I am proposing is a
evidence is provided, it would seem reason- theory about the associability of stimuli
able to suggest that a may simply be a with reinforcement. At the risk of be-
learning-rate parameter, with no effect on laboring the obvious, I must reiterate that
the control of behavior. Behavior would a is a learning-rate parameter, and pos-
then be determined by a combination of sibly a determinant of performance: There
the associative strengths of all stimuli is no implication that changes in a cor-
present and impinging on the subject's respond to changes in perception.
receptors. Theories of selective attention have all
The only set of results that may require incorporated a second assumption—that
a revision of this assumption is the ob- the probability of attending to one set of
servation, noted earlier, that a stimulus stimuli is inversely related to the proba-
imperfectly correlated with reinforcement bility of attending to others. The em-
pirical justification for this assumption has
may initially acquire associative strength
been that it predicts the occurrence of such
but will eventually lose control if it is instances of stimulus selection as over-
accompanied by other, more valid stimuli. shadowing and blocking. A closer anal-
The explanation provided by Rescorla and ysis suggests, however, that these phe-
Wagner (1972) is that such a stimulus nomena may not be due to any such
loses associative strength due to competi- competition for attention, but may rather
tion with more valid stimuli for a limited be a consequence of subjects learning to
total amount of associative strength. The ignore stimuli that signal only the occur-
present theory is better able to explain rence of a reinforcer already predicted by
why an invalid stimulus should not ac- other stimuli. This idea can be expressed
quire associative strength in the first place by the choice of appropriate rules for
changing a. Thus the single assumption
than to explain why, having acquired
that a increases and decreases in particular
strength, it should then lose it. It may ways is sufficient to explain many of the
be necessary to assume that although such data that have been thought to require
a stimulus maintains its associative a theory of selective attention.
strength, a decline in a will decrease the It is worth concluding with some quali-
probability that it will control responding. fications. It is obvious enough that the
ideas proposed here are more a program far the apparent absence of control by B
for a theory than a fully elaborated formal following reinforcement of AB reflects con-
model of conditioning and discrimination figural conditioning to the AB compound.
learning. In our present state of knowl- A second factor ignored by the present
edge, there is little need to apologize for analysis may also contribute to at least
this. It is more important to point to some cases where salient irrelevant stimuli
some factors that may limit the generality lose control of responding. From the time
of the present arguments. of their earliest discovery, results showing
First, I have assumed that it is possible apparently selective effects in conditioning
to predict the behavior of a subject toward and discrimination learning have been at-
a compound stimulus, AB, solely from tributed by some investigators to the oc-
knowledge of the status of its elements, currence of overt orienting responses. If B
A and B. The possibility of "configural" fails to acquire control over responding
conditioning or "compounding" shows that following reinforcement of AB, it is said
this is not always true: With sufficient to be because the subject oriented towards
training, animals can learn to respond ap- A and thus failed adequately to observe B.
propriately when AB is consistently rein- One of the few instances where there is
forced but A and B separately are con- actually good reason to suppose that such
sistently not reinforced (e.g., Rescorla, effects might operate is in the experiments
1972a; Woodbury, 1943). There is good of Jenkins and Sainsbury (1969), in which
reason to believe, moreover, that configural pigeons were trained to peck at visual
learning occurs even when it is not ex- displays containing discrete elements. In
plicitly reinforced. Booth and Hammond this situation, A and B would refer, for
(1971), for example, showed that con- example, to small red and green dots or
tinued reinforcement of AB might be suf- to a circle and a star; these elements were
ficient to reduce significantly the level of displayed on a large, specially constructed,
responding maintained by A and B alone, response key, which enabled the experi-
even without nonreinforcement of the com- menters to record which element was
ponent stimuli in isolation. pecked at on a given trial. If birds were
Similar configural learning is a fairly trained on a successive discrimination be-
common observation in operant discrimi- tween AB+ and B — , Jenkins and Sains-
nation learning by pigeons. If pigeons are bury found that they learned the dis-
rewarded for pecking an illuminated re- crimination by concentrating their pecks
sponse key in the presence of a 1,000 Hz at A on reinforced trials. This suggested
tone but not in its absence, they will learn the possibility that when pecking at
not to respond to the key light alone but (orienting towards) A on reinforced trials,
will equally refuse to peck in the presence the pigeon associated only A with rein-
of the tone if the key light is turned off forcement. Although nominally reinforced
(Jenkins & Harrison, 1960). This case is on 50% of trials, therefore, B lost control
of particular interest in the present con- of responding because it failed to gain
text, for it is an example of the main strength on reinforced trials. Although
unresolved problem for the analysis pro- later finding that pigeons could, to some
posed here—how a salient but relatively extent, associate B with the outcome of
invalid stimulus comes to lose control over a trial even though pecking at A, Jenkins
responding after it has initially acquired (1973) also obtained results that suggest
such control. If pigeons stop pecking that the (partial) selectivity of orienting
when the key light is turned off, it is responses in this type of situation may
obviously misleading to suggest that the indeed play an important role in enabling
light has lost control over responding. birds to learn the discrimination between
Perhaps similar configural effects occur in AB+ and B — . If in such a problem
other, similar situations. There are, in reinforcement was forfeited on positive
fact, very few data available to show how trials whenever the birds pecked at A,
they continued to peck at B on both they were clearly able to attend to both
positive and negative trials and thus failed features of the sample. But as soon as
to learn the discrimination. Because pecks exposure to the sample was reduced to a
were directed at B, it gained strength on fixed, very brief period of time, perfor-
reinforced trials and thus continued to mance deteriorated sharply, unless the
control responding. It is possible, there- subjects were given an additional, condi-
fore, that in this situation at least, the tional cue that signaled whether the en-
loss of control by the relatively invalid suing trial was going to require a line
stimulus, B, depends on a mechanism of match or a color match. The implication
selective orientation that lies outside the is that pigeons were unable to analyze
scope of the present analysis. (or remember) both features of the com-
Finally, it is necessary to acknowledge pound sample when it was presented for
the possibility that there may be situations only a brief interval. Provided the condi-
in which selective effects occur for the tional cue signified which feature they
reasons postulated by traditional theories needed to attend to, however, they main-
of selective attention. Mackintosh (1971), tained accurate performance. There was
for example, found that significant over- evidence, therefore, that stimuli were com-
shadowing of a weak tone by a strong peting for access to a limited capacity
light might occur on the very first trial of system, and the inverse hypothesis was fully
conditioning. Such overshadowing clearly supported.
cannot depend on subjects learning the Although it may seem unpleasantly
redundancy of the tone nor on the mecha- complex, it is possible that a complete
nism proposed by Rescorla and Wagner analysis of conditioning and discrimination
(1972); it suggests a perceptual or atten- learning will require the assumption that
tional interaction. It is possible that the degree of competition between stimuli
similar instances of attentional interactions may vary from one extreme, where all
would be observed more frequently if they available stimuli are analyzed on a single
were sought under more appropriate con- trial, to the other, where something like
ditions. I earlier argued that the inverse a one-look model may apply.
hypothesis of theories of selective atten-
REFERENCE NOTES
tion did not rest on any very secure
rationale: Animals may have a limited 1. Hall, G. Learning to ignore irrelevant stimuli:
Variations within and between displays. Un-
capacity for processing information, but published manuscript, 1974. (Available from
it is hard to believe that this limitation G. Hall, Laboratory of Experimental Psycho-
prevents the simultaneous analysis of the logy, University of Sussex, Brighton, England.)
relatively salient stimuli typically used in 2. Turner, C. Personal communication, 1969.
3. Mackintosh, N. J. Blocking of conditioned sup-
studies of conditioning. If there is a limit, pression: Role of the first compound trial. Manu-
however, its effects should be apparent script submitted for publication, 1975.
under suitable conditions, such as the brief
controlled presentations of complex stimuli. REFERENCES
This argument has also been advanced Booth, J. J., & Hammond, L. J. Configural con-
ditioning: Greater fear in rats to compound than
by Riley and Leith (in press), and its component through overtraining of the com-
force is suggested by the results of an pound. Journal of Experimental Psychology, 1971,
experiment, undertaken by Turner (Note 87, 255-262.
2), on delayed matching to sample in Egger, M. D., & Miller, N. E. When is a reward
pigeons. Given unlimited exposure to the reinforcing?: An experimental study of the in-
formation hypothesis. Journal of Comparative
sample key, subjects performed accurately and Physiological Psychology, 1963, 56, 132-137.
even when they did not know until after Estcs, W. K. Toward a statistical theory of
the sample had been turned off whether learning. Psychological Review, 1950, 57, 94-107.
they were going to be required to match Estes, W. K. The statistical approach to learning
theory. In S. Koch (Ed.), Psychology: A study
the color or the line displayed on the of a science (Vol. 2). New York: McGraw-Hill,
sample. Under these conditions, therefore, 1959.
Fisher, M. A., & Zeaman, D. An attention- Logan, F. A. A note on stimulus intensity dy-
retention theory of retardate discrimination namism (V). Psychological Review, 1954, 61,
learning. In N. R. Ellis (Ed.), International 77-80.
review of research in mental retardation. New Lovejoy, E. Attention in discrimination learning.
York: Academic Press, 1973. San Francisco: Holden-Day, 1968.
Foree, D. D., & LoLordo, V. M. Attention in Lovejoy, E., & Russell, D. G. Suppression of
the pigeon: Differential effects of food-getting learning about a hard cue by the presence of
versus shock-avoidance procedures. Journal of an easy cue. Psychonomic Science, 1967, 8,
Comparative and Physiological Psychology, 1973, 365-366.
85, 551-558. Lubow, R. E. Latent inhibition. Psychological
Gray, T., & Appignanesi, A. A. Compound con- Bulletin, 1973, 79, 398-407.
ditioning: Elimination of the blocking effect. Lubow, R. E., & Moore, A. U. Latent inhibition:
Learning and Motivation, 1973, 4, 374-380. The effect of nonreinforced preexposure to thi?
Halgren, C. R. Latent inhibition in rats: Asso- conditioned stimulus. Journal of Comparative,
ciative or nonassociative? Journal of Compara- and Physiological Psychology, 1959, 52, 415-419.
tive and Physiological Pyschology, 1974, 86, 74-78. Mackintosh, N. J. Extinction of a discrimination
Hara, K., & Warren, J. M. Stimulus additivity habit as a function of overtraining. Journal of
and dominance in discrimination performance by Comparative and Physiological Psychology, 1963,
cats. Journal of Comparative and Physiological 56, 842-847.
Psychology, 1961, 54, 86-90. Mackintosh, N. J. Further analysis of the over-
Hull, C. L. Principles of behavior. New York: training reversal effect. Journal of Comparative
Appleton-Century-Crofts, 1943. and Physiological Psychology, Monograph, 1969,
Jenkins, H. M. Noticing and responding in a dis- 67 (2, Pt. 2).
crimination based on a distinguishing element. Mackintosh, N. J. An analysis of overshadowing
Learning and Motivation, 1973, 4, 115-137. and blocking. Quarterly Journal of Experimental
Jenkins, H. M., & Harrison, R. H. Effect of dis- Psychology, 1971, 23, 118-125.
crimination training on auditory generalization. Mackintosh, N. J. Stimulus selection: Learning
Journal of Experimental Psychology, 1960, 59, to ignore stimuli that predict no change in rein-
246-253. forcement. In R. A. Hinde & J. Stevenson-
Jenkins, H. M., & Sainsbury, R. S. The develop- Hinde (Eds.), Constraints on learning. London:
ment of stimulus control through differential Academic Press, 1973.
reinforcement. In N. J. Mackintosh & W. K. Mackintosh, N. J., & Holgate, V. Serial reversal
Honig (Eds.), Fundamental issues in associative training and nonreversal shift learning. Journal
learning. Halifax, Canada: Dalhousie University of Comparative and Physiological Psychology,
Press, 1969. 1969, 67, 89-93.
Kamin, L. J. Temporal and intensity character- Mackintosh, N. J., & Turner, C. Blocking as a
istics of the conditioned stimulus. In W. F. function of novelty of CS and predictability
Prokasy (Ed.), Classical conditioning: A sym- of UCS. Quarterly Journal of Experimental
posium. New York: Appleton-Century-Crofts, Psychology, 1971, 23, 359-366.
1965. Medin, D. L. Subproblem analysis of discrimina-
Kamin, L. J. Predictability, surprise, attention tion shift learning. Behavior Research Methods
and conditioning. In B. Campbell & R. Church and Instrumentation, 1973, 5, 332-336.
(Eds.), Punishment and aversive behavior. New Mellgren, R. L., & Ost, J. W. P. Transfer of
York: Appleton-Century-Crofts, 1969. Pavlovian differential conditioning to an operant
Kendler, T. S. Continuity theory and cue-domi- discrimination. Journal of Comparative and
nance. In H. H. Kendler & J. T. Spence (Eds.), Physiological Psychology, 1969, 67, 390-394.
Essays in neobehaviorism: A memorial volume to Miles, C. G., & Jenkins, H. M. Overshadowing
Kenneth W. Spence. New York: Appleton- in operant conditioning as a function of discrimi-
Century-Crofts, 1971. nability. Learning and Motivation, 1973, 4,
Krechevsky, I. "Hypotheses" in rats. Psycho- 11-27.
logical Review, 1932, 39, 516-532. Newman, F. L., & Benefield, R. L. Stimulus
Lashley, K. S. Brain mechanisms and intelligence: control, cue utilization, and attention: Effects
A quantitative study of injuries to the brain. of discrimination training. Journal of Compara-
Chicago: University of Chicago Press, 1929. tive and Physiological Psychology, 1968, 66,
Lawrence, D. H. Acquired distinctiveness of cues: 101-104.
I. Transfer between discriminations on the basis Pavlov, I. P. Conditioned reflexes. London:
of familiarity with the stimulus. Journal of Oxford University Press, 1927.
Experimental Psychology, 1949, 39, 770-784. Perkins, C. C., Jr. The relation between condi-
Lawrence, D. H. Acquired distinctiveness of cues: tioned stimulus intensity and response strength.
II. Selective association in a constant stimulus Journal of Experimental Psychology, 1953, 46,
situation. Journal of Experimental Psychology, 225-231.
1950, 40, 175-188. Reiss, S., & Wagner, A. R. CS habituation
Lawrence, D. H. The transfer of a discrimination produces a "latent inhibition effect" but no
along a continuum. Journal of Comparative and active "conditioned inhibition." Learning and
Physiological Psychology, 1952, 45, 511-516. Motivation, 1972, 3, 237-245.
Rescorla, R. A. Summation and retardation tests parative and Physiological Psychology, 1966, 61,
of latent inhibition. Journal of Comparative and 198-207.
Physiological Psychology, 1971, 75, 77-81. Sutherland, N. S., & Mackintosh, N. J. Mecha-
Rescorla, R. A. "Configural" conditioning in dis- nisms of animal discrimination learning. New
crete-trial bar pressing. Journal of Comparative York: Academic Press, 1971.
and Physiological Psychology, 1972, 79, 307- Thomas, D. R. Stimulus selection, attention, and
317. (a) related matters. In J. H. Reynierse (Ed.),
Rescorla, R. A. Informational variables in Pavlo- Current issues in animal learning. Lincoln:
vian conditioning. In G. Bower (Ed.), The psy- University of Nebraska Press, 1970.
chology of learning and motivation (Vol. 6). Tighe, T. Subproblem analysis of discrimination
New York: Academic Press, 1972. (b) learning. In G. Bower (Ed.), The psychology
Rescorla, R. A., & Wagner, A. R. A theory of of learning and motivation (Vol. 7). New York:
Pavlovian conditioning: Variations in the effec- Academic Press, 1973.
tiveness of reinforcement and nonreinforcement. Trabasso, T. R., & Bower, G. H. Attention in
In A. H. Black & W. F. Prokasy (Eds.), Clas- learning: Theory and research. New York:
sical conditioning II: Current research and theory. Wiley, 1968.
New York: Appleton-Century-Crofts, 1972. Turner, C. Models of discrimination learning.
Riley, D. A., & Leith, C. R. Multidimensional Unpublished doctoral dissertation, Oxford Uni-
psychophysics and selective attention in animals. versity, 1968.
Psychological Bulletin, in press. Wagner, A. R. Stimulus validity and stimulus
selection in associative learning. In N. J.
Schnur, P. Selective attention: Effect of element
preexposure on compound conditioning in rats. Mackintosh & W. K. Honig (Eds.), Fundamental
Journal of Comparative and Physiological Psy- issues in associative learning. Halifax, Canada:
chology, 1971, 76, 123-130. Dalhousie University Press, 1969.
Wagner, A. R. Elementary associations. In H. H.
Shepp, B. E., & Eimas, P. D. Intradimensional Kendler & J. T. Spence (Eds.), Essays in neo-
and extradimensional shifts in the rat. Journal behaviorism: A memorial volume to Kenneth W.
of Comparative and Physiological Psychology, 1964, Spence. New York: Appleton-Century-Crofts,
57, 357-361. 1971.
Shepp, B. E., & Schrier, A. M. Consecutive Wagner, A. R., Logan, F. A., Haberlandt, K.,
intradimensional and extradimensional shifts in & Price, T. Stimulus selection in animal dis-
monkeys. Journal of Comparative and Physio- crimination learning. Journal of Experimental
logical Psychology, 1969, 67, 199-203. Psychology, 1968, 76, 171-180.
Spence, K. W. The nature of discrimination Waller, T. G. Effect of irrelevant cues on dis-
learning in animals. Psychological Review, 1936, crimination acquisition and transfer in rats.
43, 427-449. Journal of Comparative and Physiological Psy-
Spence, K. W. Behavior theory and conditioning. chology, 1970, 73, 477-480.
New Haven, Conn: Yale University Press, Warren, J. M., Derdzinski, D., Hirayoshi, L,
1956. & Mumma, R. Some tests of attention theory
Stettner, L. J. Effect of prior reversal and elimi- with cats. In D. Mostofsky (Ed.), Attention:
nation of inhibition on the persistence of a dis- Contemporary theory and analysis. New York:
crimination despite subsequent equal reinforce- Appleton-Century-Crofts, 1970.
ment of the discriminanda. Journal of Comparative Woodbury, C. B. The learning of stimulus pat-
and Physiological Psychology, 1965, 60, 262-264. terns by dogs. Journal of Comparative Psycho-
Sutherland, N. S. The learning of discrimination logy, 1943, 35, 29-40.
by animals. Endeavour, 1964, 23, 69-78. Zeaman, D., & House, B. J. The role of attention
Sutherland, N. S., & Andelman, L. Learning with in retardate discrimination learning. In N. R.
one and two cues. Psychonomic Science, 1967, Ellis (Ed.), Handbook of mental deficiency: Psy-
15, 253-254. chological theory and research. New York:
McGraw-Hill, 1963.
Sutherland, N. S., & Holgate, V. Two-cue dis-
crimination learning in rats. Journal of Com- (Received October 7, 1974)

A Theory of Attention. Variations in The Associability

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

A Theory of Attention. Variations in The Associability

Hochgeladen von

Copyright:

Verfügbare Formate

Psychological Review

1975, Vol. 82, No. 4, 276-298

A Theory of Attention: Variations in the Associability

In experiments on classical conditioning, response. Whatever differences there may

of increments to the associative strength tioning proceeds faster with stimulus A

surprising, therefore, to find that formal by postulating changes in a stimulus-

prevent conditioning to the second ele- strengths of all stimuli (including A)

similarly, disconfirmation of an expecta- interpretation of stimulus selection. If

we could say, Overshadowing will also be predicted if

to conditioning with that particular UCS Starting Value of a: Representation of

pigeons have established that in the CONCLUSIONS

Das könnte Ihnen auch gefallen